* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download The Cost of Immediacy for Corporate Bonds
Survey
Document related concepts
Transcript
The Cost of Immediacy for Corporate Bonds∗ Jens Dick-Nielsen Copenhagen Business School [email protected] Marco Rossi Texas A&M University [email protected] August 21, 2016 Abstract Liquidity provision in the corporate bond market has become significantly more expensive after the 2008 credit crisis. Using index exclusions as a natural experiment during which uninformed index trackers request immediacy, we find that the price of immediacy has doubled for short-term investment-grade bonds, and more than tripled for speculative-grade bonds. In addition to this level effect, after the crisis, the supply of immediacy has become more elastic with respect to its price. These results are consistent with Duffie (2012)’s prediction that the post-crisis regulatory environment would hinder market making. Keywords: Dealer inventory; Lehman/Barclay bond index; Market making; Transaction costs; Dodd-Frank Act. JEL: C23; G12 ∗ Thanks to Jennie Bai, David Blitzer, Darrell Duffie, Peter Feldhütter, Thierry Foucault, Jean Helwege, Charles Himmelberg, Soeren Hvidkjaer, David Krein, David Lando, Mads Stenbo Nielsen, Lasse Heje Pedersen, Lars Jul Overby, Jonathan Sokobin, Ilya Strebulaev, Erik Thiessen, Kumar Venkataraman, Avi Wohl, and participants of the Annual Fixed Income Conference in Charleston (2012), Midwest Finance Association conference in Chicago (2012), the FINRA and Columbia conference on Corporate Debt Market Structure, Liquidity, and Volatility (2015), the 2016 4Nations cup (1st place), the Corporate bond workshop at EM Strasbourg (2016), the Financial Econometrics and Empirical Asset Pricing Conference in Lancaster (2016), and the Financial Conduct Authority (London), the Danish CFA society, Copenhagen Business School, SMU, Texas A&M, and Notre Dame seminars for their helpful comments. The paper is scheduled for presentation at the 2016 Notre Dame regulation conference, the 2016 MIT conference on regulation and the credit crisis, and the 2017 AFA in Chicago. Special thanks to FINRA and Jonathan Sokobin for providing the transaction data and for discussions regarding the data. Also special thanks to Charles Himmelberg from Goldman Sachs for providing the aggregate corporate bond inventory data. The paper won second prize in the 2012 SPIVA awards for best index application. Jens Dick-Nielsen gratefully acknowledges support from the Center for Financial Frictions (FRIC), grant no. DNRF102. and from the Danish Social Science Research Council (FSE). 1 Introduction In dealership, over-the-counter markets dealers are compensated for providing liquidity to market participants. Dealers provide liquidity in the form of immediacy by expanding, or reducing, their inventory to facilitate trading, as opposed to just matching customers (Garman, 1976; Stoll, 1978; Amihud and Mendelson, 1980; Ho and Stoll, 1981). Since the onset of the 2008 credit crisis, aggregate corporate bond inventories have shrunk by 50-60%.1 During the same period the corporate bond market has been growing steadily: the amount of corporate bonds outstanding has almost doubled; trading volume has increased; and the number of new issuances has increased (see e.g. Figure 1 panel C and D). Shrinking inventories and a growing bond market suggest that providing immediacy to liquidity seekers has become harder. We show that the price of immediacy has more than doubled relative to what it used to be before the 2008 crisis. Following the credit crisis, dealers have kept their inventories low in anticipation of regulation banning proprietary trading and imposing tighter capital requirements, (Watson, 2012). These regulatory responses to the crisis were already apparent by early 2009, and the precise formulations have long been debated and eventually implemented. While regulation could have the unfortunate side-effect of reducing market liquidity (Duffie, 2012), empirical investigations based on effective bid-ask spreads, or other price impact measures, are unlikely to uncover its adverse effect on liquidity.2 A direct application of the well-known Lucas (1976) critique, suggests that, by increas1 Primary dealer inventories of corporate securities peaked at the beginning of the crisis at $281 bn. In early 2009 the inventories had been cut to $100 bn and in September 2012 the inventories were down to $60 bn (see Figure 1). The specific drop in inventories of corporate bonds is also approximately around 50% from the peak before the crisis. 2 For instance, Trebbi and Xiao (2015); Adrian, Fleming, Shachar, and Vogt (2015); Bessembinder, Jacobsen, Maxwell, and Venkataraman (2016) all find some evidence that liquidity measures and trading costs have improved after the crisis. 1 ing the cost of immediacy, the regulation may affect the optimal behavior of market participants, who will likely, albeit reluctantly, spread their large trades over time, or give up on large trades altogether. In fact, measured trading costs might decrease, leading to erroneous conclusions about recent regulations. To use an analogy, rules increasing the cost of air travel will induce more travelers to use the bus. By discouraging air travel, such regulation might well lower the average cost of transportation (after all, taking the bus is cheaper than a plane ticket), but average utility will surely decline because of the loss of immediacy. Getting from Los Angeles to New York in three days by bus is not the same as completing the trip in five hours by plane. Liquidity is generally defined as the ability to execute a transaction at a fair price and on short notice. Low realized bid-ask spreads thus indicate that transactions are executed near a fair price, but they tell little about the speed of execution, i.e. whether immediacy was supplied by the dealer. Since post-crisis regulation might have impacted the ability to provide immediacy, we need to look beyond traditional corporate bond liquidity measures. The main contribution of this study is to quantify the cost of immediacy for corporate bonds in a trading environment that circumvents the Lucas (1976) critique. We identify trading situations in which the motive to obtain immediacy is strong, so that liquidity seekers cannot orchestrate alternative trading arrangements. Furthermore, in our setting, the desire to trade reveals no information about the fundamental value of the assets traded. Specifically, we compute liquidity costs around bond exclusions from the Barclay Capital (formerly Lehman) investment-grade corporate bond index. In this natural experiment, index trackers (the sellers) request immediacy from the dealers (the buyers) in order to minimize their tracking error. Moreover, mechanical index rules, not fundamentals, dictate the decision to trade, so that dealers do not have to worry about information motivating the trades. This last observation ensures 2 that the dealer’s bid reflects the cost of providing immediacy, rather than the adverse selection problem of dealing with unwanted informed traders (Easley and O’hara, 1987). Our empirical analysis shows that the price elasticity of the supply of immediacy has increased significantly after the crisis. This increase in elasticy is indicative of higher market making cost, which translates into higher average transaction costs,3 thus providing support for standard theories of market maker inventories. For safe bonds, which are quickly turned over again by dealers, the cost of immediacy has approximately doubled, while for more risky bonds, the cost has more than tripled. The adverse increase in transaction costs offsets the impact of previous regulations in the corporate bond market that aimed at lowering the cost of trading through increased transparency (Bessembinder, Maxwell, and Venkataraman, 2006; Edwards, Harris, and Piwowar, 2007; Goldstein and Hotchkiss, 2008). We also show that the market share of traditional dealers decreases substantially after the crisis, as predicted by Duffie (2012). Hence, as it becomes more expensive for traditional market makers to supply immediacy other market participants can take over as market makers but at a higher cost. We infer the cost of immediacy by computing an inter-temporal bid-ask spread, which we define as the percentage difference between the post-exclusion ask price and the pre-exclusion bid price. This measure captures the essence of the dealer’s role, who uses her inventory to absorb the selling pressure generated by the index trackers unloading their positions, and then resells the bonds to restore the desired level of inventory. In order to control for systematic movement in the corporate bond market around the exclusion, we also compute a measure of abnormal bond performance using 3 On average, the estimated bid-ask spreads in this study are comparable to those of Edwards, Harris, and Piwowar (2007) and Feldhütter (2012). 3 the methodology proposed by Bessembinder, Kahle, Maxwell, and Xu (2009). Finally, we use a special version of the TRACE database, which we obtained directly from FINRA, to compute dealer specific returns and bond positions around the exclusions. All these measures point to the same conclusion that the cost of providing immediacy has increased in the post-crisis, low-inventory regime. Before measuring the transaction costs around index exclusions, we verify that these exclusions are indeed events during which index trackers request immediacy. Our analysis reveals that the traded volume of bonds exiting the index peaks during the day of the exclusion, and it is at least five times higher than in the weeks surrounding the exclusion. The peak in trading volume is consistent with index trackers attempting to minimize their tracking errors by trading close to the index exclusion date. Blume and Edelen (2004) show that stock index trackers display a similar desire to transact on the exact exclusion date. Having established the existence of a demand for immediacy, we verify that dealers absorb the selling pressure by increasing their inventories. Dealers slowly increase their inventories several days before the exclusion, with the increase intensifying at the exclusion day. The speed with which bonds are resold by the dealers in the weeks following the index exclusion depends on the nature of the exclusion: for maturityrelated exclusions, the sales take place relatively quickly; for downgrade-related exclusions, the bond sales tend to occur over a longer period. Dividing the sample into three sub-periods shows that dealers’ behavior has changed after the crisis. Our analysis of the cumulative change in inventories demonstrates that dealers’ willingness to hold the bonds in their inventories has declined in the new regulatory regime. While before (and even during) the crisis dealers kept most of the downgraded bonds for at least one hundred days, after the crisis the inventories return to pre-exclusion levels within 20 trading days. Risky bonds are no longer kept 4 on inventory thereby presumably reducing the riskiness of the market maker. This reduction in risk-taking by dealers could be seen as a success of the Volcker Rule. For bonds maturing within less than one year, dealers start selling soon after the exclusion date, and inventories are back to normal within two weeks. We note that, during the crisis, the cumulative change in inventories for the short-maturity bonds becomes negative within a day of the exclusion, probably because selling these liquid bonds was the easiest way to raise cash. We also note that the dealer behavior for the downgraded bonds were very similar before and during the crisis where a large part of the bonds stayed on the dealers inventory. It is after the crisis, rather than during the crisis when dealers should have been more constrained, that the dealer behavior changes. This change in dealer behavior suggests that it is driven by regulation concerns, rather than limited risk bearing capacity in a more traditional sense. In addition to contributing to the literature on corporate bond liquidity, this paper occupies a natural place in the literature connecting regulations to financial market efficiency. The debate on the repercussions of the Dodd-Frank act on the financial system offers positions that view the regulatory changes as harmful (Duffie, 2012) as well as beneficial (Richardson, 2012). Despite the multitude of opinions on the welfare effects of the regulatory changes that took place after the credit crisis, few empirical investigations exist. In a recent paper, Trebbi and Xiao (2015) argue that the Dodd-Frank act has not had negative welfare consequences, since their breakpoint testing technique cannot detect any “breakpoints in market liquidity for fixed-income asset classes and across multiple liquidity measures, with special attention given to the corporate bond market.” However, we have already argued that these liquidity measures (such as the one shown in Figure 1, Panel B) are the outcome of market participants’ optimization problems, and a large-scale policy change alters the rules of the game, that is the parameters of that optimization. By focusing on a homogenous, 5 information-free event in which agents cannot arrange alternative trading strategies both before and after the policy change, our analysis is able to uncover the adverse effect that the new regulatory regime has had on corporate bond liquidity. Bao, O’Hara, and Zhou (2016) confirm our findings regarding the impact of the Volcker Rule using a more recent sample. Consistent with our study, Bessembinder, Jacobsen, Maxwell, and Venkataraman (2016) provide some evidence that dealers are less willing to commit capital after the crisis. Our paper also contributes to the literature on index revisions, which is quite large when it deals with stock index revisions (Harris and Gurel (1986); Shleifer (1986); Beneish and Whaley (1996); Hegde and McDermott (2003); Denis, McConnell, Ovtchinnikov, and Yu (2003); Chen, Noronha, and Singal (2004); Greenwood (2005)). Bond index revisions have also been studied in Newman and Rierson (2004) and Chen, Lookman, Schürhoff, and Seppi (2014), but both studies have focused on special one-time announcement effects, months before the actual index revision date. Newman and Rierson (2004) look at a large and unique issuance event for European telecom companies. Chen, Lookman, Schürhoff, and Seppi (2014) look at the effect of a unique rating rule change for the Lehman index. Different from those studies, the empirical part of this paper looks at the trading very close to the actual index revision dates. Lastly, the finding of our paper have bearing on the literature studying liquidity provision around predictable trades. Bessembinder, Carrion, Tuttle, and Venkataraman (Forthcoming) show that traders engaged in predictable transactions, that do not reveal information about the fundamentals of the traded asset, need not face predatory trading (Brunnermeier and Pedersen, 2005). In fact, disclosing information about the timing and attributes of the transaction mitigates its adverse price impact (Admanti and Pfleiderer, 1991). Our results indicate that the finding of Bessem6 binder, Carrion, Tuttle, and Venkataraman (Forthcoming) on the periodic portfolio re-balancing by USO (a commodity ETF) are valid in a more general market setting, even when trades are not completely predictable. Lou, Yan, and Zhang (2013) look at anticipated shocks in the treasury market whereas our study focuses on the more illiquid corporate bond market. 2 Corporate bond index tracking Exclusions from the Barclay Capital (formerly Lehman) corporate bond index provide an ideal natural experiment for studying the cost of immediacy over time. Each month corporate bond index trackers demand immediacy from dealers when they seek to sell bonds exactly when the bonds exit the index. The rules for bonds entering or exiting the index are both transparent and mechanical which makes the monthly exclusion events information-free and homogeneous over time. Bonds exit the index for three main reasons: time to maturity falls below 1 year; their rating goes from investment-grade to speculative grade; or when they are called by the issuer. The index is rebalanced once a month on the last trading day of the month at 3:00 PM EST and all bonds that are no longer index eligible are excluded at this point in time. Hence, the actual downgrade date of a bond which may or may not contain information (Ambrose, Cai, and Helwege (2012)) takes place before the bond exclusion from the index (we return to this issue in the robustness section). The objective of the index trackers is to minimize tracking error between the return on their portfolio and that of the index. Blume and Edelen (2004) show that index trackers following the S&P 500 index are transacting on the exact day that the index is rebalanced, even though they then sacrifice potential profit by doing so 7 (Beneish and Whaley (1996)). Low tracking error is a signal to investors that the index tracker is in fact committed to tracking the index and thus resolves an agency problem. Figure 2 shows that corporate bond index trackers, like the S&P 500 index trackers, also seek to transact as close as possible to the exclusion date. Day 0, the event day, is the last trading day of the month in which the bond is excluded. Panel A of Figure 2 shows trading volume for all bonds excluded from the index because of low maturity. Trading volume is aggregated across all the bonds excluded at a given event date and then averaged across all event dates. Panel B replicates Panel A for bonds excluded because of a recent downgrade to speculative grade. In both cases, trading activity spikes at the exclusion date.4 Since corporate bonds trade over-the-counter, index trackers cannot be certain to transact at the desired date which is why activity is also high right before and after the exclusion date. Figure 2 clearly confirms that some investors are in fact tracking the index and that they seek to minimize their tracking error as is expected from index trackers. Bond index trackers are different from stock index trackers in the way they track the target index. Stock index trackers use an exact-replication strategy (Blume and Edelen, 2004), whereas bond index trackers use a sampling strategy (Schwab, 2009; Vanguard, 2009). Exact-replication implies that the investor holds a position in each asset member of the index. For corporate bond index trackers, such a strategy would generate huge transaction costs because the index is large, the market is illiquid, and the index is rebalanced every month. Instead, bond index trackers sample the index by holding only a fraction of the bonds currently in the index. This portfolio is designed to match the index with respect to duration, cash flows, quality and callability. As 4 A similar trading pattern can be seen around revisions of the S&P 500 (Shleifer (1986); Harris and Gurel (1986); Chen, Noronha, and Singal (2004) and others), the Nikkei 225 (Greenwood, 2005) and the FTSE 100 (Mase, 2007). 8 an example, the Vanguard Total Bond Market Index Fund held 3,731 out of 9,168 bonds in the Barclay Capital US aggregate bond index on December 31, 2008. All the large bond index funds, e.g. Vanguard, Schwab and Fidelity, have similar guidelines for tracking an index by sampling. The typical rule is to have 80% for their assets invested in bonds currently in the index and the remaining 20% invested outside the index. The outside investments are usually in more liquid instruments such as futures, options and interest rate swaps but could also be in non-public bonds or lower rated bonds. The 80% rule followed by index funds is not per se binding but still gives yet another reason why index trackers are selling out close to the exclusion date as we saw in Figure 2. Lehman has also given guidelines on how to track the index using only liquid derivatives (Dynkin, Gould, and Konstantinovsky, 2006) but that is not the way bond index funds are tracking the index. The criteria for how to invest the last 20% outside the index are rather loose (Schwab, 2009; Vanguard, 2009) so it is not possible to know exactly which assets the funds have on their balance sheets. The lack of transparency makes it even more important for the funds to keep a low tracking error as a way to signal sane investments (Blume and Edelen, 2004). Looking at the Vanguard Total Bond Market Index Fund again, they have had an yearly average absolute return tracking error on their shares compared to the target index of 23.5 bps over 1995-2009. This track record can be compared to that of Barclays Global Investors fund that tracks the S&P 500 index with a tracking error of only 2.7 bps per year (Blume and Edelen, 2004). 9 3 Data This study uses a unique and complete transaction level dataset for US corporate bonds provided to us by FINRA. The dataset is identical to the Enhanced TRACE dataset available on the Wharton Research Data Services (WRDS), except that we also have anonymized counterparty identifiers for each transaction. This allow us to track the changes in individual dealer inventories around the exclusion events. 3.1 Sample Construction The bond sample consists of all bonds exiting the Barclay Capital corporate bond index (formerly the Lehman corporate bond index) between July, 2002 and November, 2013. The exclusions are fairly equally scattered over time as seen in Figure 3. As of July, 2005, the index contains all US corporate bond issues with an investmentgrade rating by at least two of the three major rating agencies (Standard and Poor’s, Moody’s and Fitch). Furthermore, the issuance size must be at least $250 millions and time to maturity must be above 1 year5 . Bonds leave the index for the three reasons mentioned earlier: they are excluded if the time to maturity falls below 1 year: if they are called: or if the index rating6 goes from investment-grade to speculative grade. On the other end, bonds enter the index for two main reasons: if they are newly issued and index eligible; or if the index rating goes from speculative grade to investment-grade. These rules result in an index which covers a very large fraction of the entire market. 5 There are certain more qualitative rules for being index eligible. See index rules at https://ecommerce.barcap.com/indices/index.dxml 6 Each bond has an index rating defined as the middle rating from Standard and Poor’s, Moody’s and Fitch. If only two ratings are available, the lower and more conservative rating is used. If only one rating agency provides a rating, then that is the index rating. Before July 1st, 2005 Fitch was not used in the index rating. 10 Table 1 gives characteristics of the excluded bonds. A large number of bonds have been excluded from the index for ”other” reasons. The average issuance size of these bonds is far less than for the rest of the sample. Most of these exclusions have been generated by an increase over time in the lower index limit on issuance size, once in October, 2003 and again in June, 2004. At the end of 2009 the index contained more than 3,400 bonds. The TRACE data is cleaned up before usage following the guidelines in DickNielsen (2009). We then remove residual price outliers as in Rossi (2014). Finally, we only keep trades equal to, or above $100,000 in nominal value (Bessembinder, Kahle, Maxwell, and Xu, 2009).7 3.2 Intertemporal bid-ask spreads Calculating corporate bond returns is a challenge because most bonds trade infrequently. We calculate dealer-bond specific returns and as a robustness check, we also calculate bond specific returns. Dealer-bond specific returns are calculated by first calculating a dealer specific buying price for each bond. The dealer specific buying price is the average buying-volume weighted price over day -2, -1, and 0 for a given bond. Here day 0 is the index exclusion day, and day -2, and -1 are the two days leading up to the event date. The price thus reflects the individual dealer’s average buying price. Second, bonds may not have had transactions on all days surrounding the event date. To circumvent this problem, we calculate a market-wide average dealer-selling price on each day following the event date. The selling price is the volume weighted average selling price over all trades and dealers in that bond. Since this calculated selling price can be seen as a market-wide price, it is likely the price that the individual dealer would use to mark-to-market her acquired inventory position. 7 We keep agency transactions, but results in this study are robust to their exclusion. 11 The return is then calculated as the logarithmic difference between these two prices (Cai, Helwege, and Warga, 2007; Ambrose, Cai, and Helwege, 2012). If there are no transactions on a given day following the event date the return is calculated using the first available price after that date. In order to limit any information bias caused by the non-trading days, the sample is restricted to bonds where the prices are observed within three days of the non-trading date.8 The returns are also corrected for any accrued interest. Third, the abnormal return is formed by subtracting the return of a benchmark index (Barber and Lyon, 1997). The benchmark is a portfolio of bonds matched on rating and time to maturity. In order to test the robustness of the dealer-bond specific return, we also calculate a bond specific return. This is done using a procedure similar to the one above. The difference is that instead of a dealer-bond specific buying price, we use a market-wide buying price. The market-wide buying price is the volume weighted average across all buy-side trades and all dealers in that bond. The cost of immediacy is here defined as the return on the transaction as seen from the dealer’s viewpoint. Therefore the bid-ask spread is included in all returns as explained above. The before-price is calculated using only buy-side prices and the after-price is calculated using only sell-side prices. This method resembles the return to the dealer who first buys the bonds from the index trackers and then sells them on to other investors. We note that these returns are not easily replicable by other investors in the economy, who would face a possibly large bid-ask spread to implement the strategy of buying at the exclusion and selling afterward. The rest of the study uses the following terminology. When the benchmark return is subtracted from the raw return it is called an abnormal return, and when the benchmark return is not 8 Both Cai, Helwege, and Warga (2007) and Ambrose, Cai, and Helwege (2012) show that this return calculation method is robust to an even larger window. 12 subtracted it is called an inter-temporal bid-ask spread. The latter method is also used as the event return in Goldstein and Hotchkiss (2008), whereas the the former method is used as the event return in Cai, Helwege, and Warga (2007) and Ambrose, Cai, and Helwege (2012). 4 Event study of index exclusions 4.1 Dealer inventory Index trackers demand immediacy when bonds exit the index. This section shows that dealers take the bonds on inventory when providing immediacy. Since the dealers’ attitude towards taking on new inventory shifts over time, it is expected that the cost of immediacy for the index trackers also shifts - consistent with inventory theories. Figure 2 showed that trading activity spikes at the exclusion events indicating that index trackers get rid of the bonds. Figure 4 shows the corresponding dealer inventories in the bonds excluded because of low maturity and because of a recent downgrade. The inventories are cumulative, aggregated over all dealers, and with a chosen benchmark of $0 100 trading days before the event. The daily change in inventory is calculated as the total volume in dealer buys minus the sales. Consistent with the story of index trackers demanding liquidity or immediacy at the exclusions, the dealer inventories on average increase in the days leading up to the exclusion and particularly on the event day. Hence, dealers use their inventory actively when providing liquidity. For the low maturity bonds, we see the increase starting around 3 days prior to the exclusion date whereas the buildup for the downgraded bonds starts earlier but also increases in magnitude approximately 3 days prior to the event. The buildup in the downgraded bonds from day -23 up to day -4 is caused by a buy 13 up from the dealers on the actual downgrade date. On the downgrade date itself other investors, different from index trackers, demand liquidity because many firms have an investment policy that discourages holding speculative grade assets. This sell out on the downgrade date happens despite a grace period of up to two month in which the institutional investors are allowed to hold on to these bonds (see e.g. Ellul, Jotikasthira, and Lundblad (2011) and Ambrose, Cai, and Helwege (2012)). However, we will show later that in terms of immediacy, the downgrade date is a smaller event than the exclusion date. At the downgrade date, dealers behave more like brokers, matching buy and sell orders, rather than providing immediacy by taking the bonds on inventory. After the exclusion event, Figure 4 shows that the dealers sell all or part of their newly acquired inventory. After 2 weeks the inventory of the low maturity bonds has been sold off. For the downgraded bonds only 60% of the bonds has been sold after 100 days. The two events thus differ in the way dealers use their inventory. Since dealers on average do not sell one-third of the buildup again within 100 days, the decrease in the general willingness to hold inventory is expected to have affected the transaction cost of the downgraded bond the most. 4.2 Dealer behavior before and after the credit crisis Figure 5a and 5b show the change in dealer inventories around the event before, during, and after the crisis. The before period is from 2002Q3 to 2007Q2, the crisis period is from 2007Q3 to 2009Q4, and the after period is from 2010Q1 to 2013Q4. The dealer behavior in the short maturity bonds has changed from before to after the crisis in that the dealers on average provide twice as much immediacy after the crisis than before. But they decrease the inventory to 0 over the same time interval. 14 Hence, the absolute speed with which they sell off again has doubled. The fact that dealers provide more immediacy after the crisis could be due to index tracking becoming more popular. If index tracking has become more popular over time then the dealer-buying volume (i.e. index trackers selling volume) should have increased over time for the regular maturity exclusions. However, this is not what we see in the data. While index trackers sell slightly more bonds at the exclusion, this cannot alone explain the increase in inventory positions that we observe. Instead, it seems that dealers’ reduced matching of buy and sell orders is responsible for the larger inventory buildup. For the downgraded bonds there is a clear shift in dealer behavior from before and during the crisis to after the crisis. Before and during the crisis the dealers keep a large fraction of the inventory increase on their books. However, after the crisis they decrease the relative position to 0 after around 21 days. Since the shift in behavior happens after the crisis, and not during the crisis, it is reasonable to infere that the shift is not driven by limited risk-bearing capacity by the dealers. Measures of dealer risk-bearing capacity such as dealer leverage, or the VIX index are all better after the crisis than during the crisis. Since the shift cannot be motivated by financial constrains it seems likely that the shift is due to anticipated tighter regulation. After the crisis aggregate dealer inventories have stayed low and we can see the same tendency in our graphs. This change in behavior indicates that the Volcker Rule has been successful in discouraging market makers from keeping a risky inventory portfolio. The risky downgraded bonds are no longer kept on inventory but are instead unloaded rather quickly. This change in behavior has most likely decreased the riskiness of the market makers as was the intention of the post-crisis regulation. Note that much of the regulation is not finally implemented during our sample period. However, the change in behavior happens before the actual implementation date 15 because most of the regulation can be anticipated and because the dealers will not risk ending up with a large inventory position which they then have to unload after the implementation of the various rules. 4.3 The cost of immediacy Since dealers actively use their inventories to provide liquidity to index trackers, we expect them to earn a positive return on average. The following section shows that dealers are compensated for providing liquidity which indirectly is the cost of immediacy for the index trackers. The costs are higher for the downgrade event compared to the low maturity event as would be expected, since the downgraded bonds are both more risky and kept longer on inventory. Table 2 and 3 show the dealer returns for the two exclusion events. Returns are either inter-temporal bid-ask spreads, or abnormal dealer returns. Each of these returns are either equally weighted or value-weighted. The value-weighted returns are weighted by the dealer buying volume (VW1), or by the dealer inventory buildup (VW2), on the event date and over the previous two days. Hence, those bonds purchased by dealers that increased inventory – provided immediacy – are given more weight. The bonds where the inventory build up is negative (because a dealer sold more bonds than she bought) are given a weight of 0. The reader should note that the inter-temporal bid-ask spreads (and the abnormal returns) are dealer and bondspecific. The price at time zero refers to the volume-weighted average purchase price that a dealer paid over the two days preceding and including the actual exclusion date. The final selling price is the market selling price of a given bond at time t, where t is the upper bound of the event window, and is common to all dealers. Looking at Table 2, we see that the abnormal dealer returns for the bonds excluded 16 due to maturity are uniformly higher after the crisis relative to before the crisis. The equally-weighted column (EW) attributes the same weight to all excluded bonds, even those for which there is very little trading, or inventory build-up. Given the statistical sampling approach to replicating the index, indexers only hold some excluded bonds. For this reason, equally weighted returns may mistakenly give too much weight to bonds for which traders do not seek immediacy. Both value-weighted returns show a much sharper increase (roughly a 100%) in the cost of immediacy for highly rated, short-term bonds over the sample period. Figure 6a, 6c, and 6e illustrate the increase in the cost of immediacy documented in Table 2. Qualitatively, Table 3 paints the same picture as Table 2. Quantitatively, the returns are much larger, which is to be expected given the low rating of these bonds and the increased inventory risk that they pose. Moreover, the increase in the cost of immediacy since the pre-crisis period is much larger than the maturity exclusion case. As can be seen from the value-weighted abnormal return columns, the increase ranges from 200% at the one-day horizon to almost 700% at the 30-day horizon. Figure 6b, 6d, and 6f illustrate the increase in the cost of immediacy documented in Table 3. The returns that we use to construct Table 2 and 3 are dealer and bond specific, meaning that we have multiple dealer returns for each bond, depending on the purchase price at which a bond entered a given dealer inventory. While this approach is not typical, it is valuable since it provides a richer picture and will help us control for dealer-specific effects in the multivariate analysis that will follow. Nevertheless, in Table 4 and 5 we replicate the event study using bond returns constructed with just an initial buying price and a final selling price. These prices are just the volume-weighted average of the transacted prices in a given day. As can be seen, the conclusions that we can draw from these tables are very similar to those we have drawn so far. Therefore, going forward, the analysis will use the finer dealer-bond-specific return structure. 17 In addition to increasing the cost of immediacy, the post-crisis regulatory regime is also expected to change the composition of market makers (Duffie, 2012). We find support for this hypothesis by looking at the market share for traditional dealers. We define market share as the fraction of the total dealer buying volume on day -2, -1 and date 0 absorbed by each dealer. Table 6 shows the combined average market share for four large market makers over time. The market makers are the four most active dealers who were alive and active in the entire sample period from 2002 to 2013. On average these dealers have a total market share before the crisis of 21% of the maturity exclusions and 32% of the downgrade exclusions. During the crisis these dealers apparently became very constrained and lost a lot of their market share. After the crisis these dealers have failed to recover their original market share. The loss in market share happens despite other very active dealers defaulting during the crisis and thus leaving a market making void. The most active dealers before the crisis are most likely subject to the Volcker Rule and other capital requirements. Consistent with the predictions in Duffie (2012), these dealers are constrained in the post-crisis regulatory regime, making it possible for other market participants to move in and compete with traditional market makers. 5 Regression analysis of the cost of immediacy In Section 4, we have shown a remarkable increase in the price of immediacy (proxied by abnormal dealer returns) taking place since the onset of the credit crisis. In this section, we relate the higher returns earned by dealers to the quantity of bonds transacted, and other variables likely to affect the supply and demand of immediacy. 18 5.1 Setup The price (P) and quantity (Q) of immediacy are jointly determined in the market. Therefore regressing the compensation for immediacy on its quantity subjects the econometrician to simultaneous equation bias. Importantly, we do not usually know if such regression estimates a supply function or a demand function. More formally, the empirical model to consider is given by the system of equations: QD = α0 + α1 Pt + et t (1) QSt = β0 + β1 Pt + ut (2) QD = QSt = Qt , t (3) where et , ut contain both observable and unobservable supply shifters, and the last equation imposes market clearing. In order to obtain unbiased and consistent estimates of the slopes, a two-stage least squares (2SLS) is normally used. See Choi, Getmansky, Henderson, and Tookes (2010) for a recent application to this methodology to the analysis of issue proceeds and underpricing for convertible bonds. The premise of this study is that indexers are impatient around bond exclusion events. Our empirical analysis so far suggests that their price demand elasticity around these events is extremely low. Therefore, the exclusion restriction that we impose is to set α1 = 0 in Equation (1). Chacko, Jurek, and Stafford (2008) impose a similar restriction in their theoretical model of the price of immediacy, but in the context of a limit order book. In summary, setting α1 = 0 identifies the relation between prices and quantities as a supply relation. A non-negative relation between prices and quantities in our setting would provide support for our assumption. 19 5.2 Model specification The dependent variable in the regressions is the cumulative abnormal bond returns (a proxy for the cost of immediacy, i.e. P(Q)). The independent variable of interest is a measure of liquidity provision (Q). Assuming that dealers see the excluded bonds as reasonable substitutes, we define Q as the aggregate dealer inventory imbalance (measured in $MIO) for each dealer from day -2 to 0 across all excluded bonds at the event (downgrade and maturity separately). We drop all dealers with a net negative inventory imbalance. We interact Q with three dummies indicating whether the observation takes place before, during, or after the credit crisis. As a robustness check, in Section 6 we will consider alternative specifications that adopt different definitions of both P and Q. We also include other factors likely to influence the cost of immediacy. Specifically, we include the amount outstanding of the bond. Larger bonds are likely more transparent and liquid than smaller bonds and are therefore less risky to have on inventory. We also include a range of variables which proxy for dealers’ risk bearing capacity. These are the VIX index as in Lou, Yan, and Zhang (2013) and aggregate leverage growth for broker-dealers from the Federal Reserve Flow of Funds data9 as in Adrian, Etula, and Muir (2014). The VIX is supposed to proxy for the dealers’ funding constraints and the aggregate leverage growth is supposed to capture the dealers’ leverage constraints. When aggregate leverage growth is low it has become more costly to obtain leverage and when VIX is high dealers face higher funding constraints. We also include the TED spread to proxy for money market stress. Finally, we include industry (financials Vs non-financials), rating, and period dummies that are interacted with each other. Moreover, we also include dealer fixed effects. To save 9 This data is for primary dealers in the treasury market. Schultz (2001) shows that the major corporate bond dealers overlap significantly with the primary market dealers in the treasury market. 20 space, the estimated coefficients on the dummies are not reported in the regression tables. Table 7 provides descriptive statistics on the variables used in the regression analysis. In addition to pooled statistics (Panel A), the table provides statistics grouped by exclusion event. Given the index inclusion criteria, the average size of the bond index is quite large. Looking at the dealer inventory buildup (Q) in Panel A, it can be seen that the average dealer inventory buildup was about $7.89 MIO before the crisis, $2.85 MIO during the crisis, and then after $6.25 MIO. Looking at Panel B and C, we see that the recovery in the size of the provision of liquidity is due mostly to maturity exclusions. The average size of Q is higher after the crisis for short-term highly rated bonds ($6.63 MIO Vs $4.88 MIO). On the contrary, for downgraded bonds, we go from an average Q of $11.98 MIO to an average Q of $5.10 MIO. These differences in average Q suggest a migration away from the risky business of providing liquidity for junk bonds toward the safer business of providing liquidly for basically money-market instruments. In this regard, Dodd-Frank might be seen as a successful regulation. 5.3 Cost of immediacy before and after the crisis Table 8 reports the coefficient estimates of the regressions. As can be seen from the table, the price of providing liquidity is increasing in the amount of liquidity transacted, making the relation reminiscent of a supply curve. Comparing the interaction of Q with the post-crisis dummy to the interaction of Q with the pre-crisis dummy reveals that the supply curve is relatively steeper after the crisis. This result suggests that providing immediacy has become more costly after the crisis, and, consequently, dealers require higher returns for providing additional immediacy. The results on the effect of Q in Table 8 are also economically significant. Noting 21 that the returns are measured in basis points, a one standard deviation change in Q after the crisis ($17.73 MIO) is associated to roughly an additional 18 basis points of return over three days, and roughly 37 (17.73 × 2.17) basis points over a 30 day horizon. Ultimately, the money to compensate dealers for their provision of liquidity comes from the final investors, not the indexer, in the form of lower returns. The control variables behave as expected although they are not always significant. Larger bonds have a lower cost of immediacy. When dealers are more constrained, i.e. a low leverage growth or a high VIX, the cost of immediacy is higher. Also when the money market undergoes stress, i.e. a high TED spread, the cost of immediacy increases. Money markets are important for market makers since they often fund their market making activity through repo transactions. 6 Robustness checks 6.1 Are downgrades and downgrade exclusions separate events? The returns for the downgraded bonds could in theory be affected by a slow postdowngrade price adjustment (Katz, 1974; Grier and Katz, 1976).10 In order to verify that this is not the case we look at dealer behavior at the actual downgrade date. Figure 7a and 7b show that the downgrade date itself also sees a lot of trading activity. However, the average turnover is of the same size as that on the exclusion date. Also, the inventory build up on the downgrade itself is far smaller than that on the exclusion date. Where inventory peaked at the exclusion date and then decreased, here the peak is delayed. The delay is consistent with an inventory build up at the exclusion date. 10 See Norden and Weber (2004) for a review of other rating change studies 22 Figure 8 shows both the downgrade date and the exclusion date for events with a fixed number of days between the exclusion and the downgrade. The fixed intervals of 4, 11, and 17 days are chosen because these are the days with most turnover on the downgrade date. The volume figures clearly show two spike in trading activity, first on the downgrade date and then on the exclusion date. The inventory graphs show that there sometimes is an inventory increase at the downgrade date but that there always is a second large increase at the exclusion date. After this second spike the inventory immediately starts to decrease. In summary, the analysis verifies that the downgrade event and the exclusion event are two separate events. There may be information on the downgrade date itself, but there should be little or no information on the exclusion date. Also note that the trading activity for the downgrade date is also highly concentrated on the downgrade itself. The downgrade event is thus fairly short lived. The downgrade date is not well suited for investigating the cost of immediacy because of the potential information contained in the event and because dealers provide little immediacy using their inventory compared to the exclusion date. 6.2 Alternative regression model specification In relating prices to quantity, we have proxied the price/cost of immediacy with dealers’ abnormal returns, and the quantity of immediacy with dealers’ inventory buildup, measured in millions of dollars. In this section, we explore alternative specifications for both the price and quantity of immediacy. Table 9 presents coefficient estimates of a regression relating abnormal returns to the natural logarithm of inventory buildup. The coefficients on the variable Q can now be interpreted as elasticities. As can be seen from the table, the coefficient on Q 23 (measured in natural logarithm) is much larger after the crisis. Table 10 presents coefficient estimates of a regression in which we proxy for the price of immediacy (the dependent variable) with the actual bond price paid by dealers at the exclusion event. Specifically, the dependent variable is the current price (the denominator) used to calculate the inter-temporal bid-ask spreads and abnormal returns. One of the benefits of this specification is that it does not require a final selling price to compute returns, thus leading to a larger sample size. One of the drawbacks of this specification is that prices are a bit more unstable than returns. To better account for differences in bond price levels, we also include the coupon rate and the time to maturity as additional controls. As can be seen from Model 2 of Table 10, the coefficient on Q (measured in $MIO) is much more negative after the crisis, and it is the only one that is actually significant. Note that, since the dependent variable is a price, rather than a return, a negative coefficient means that a dealer requires a higher expected return (lower price today) in order to provide additional liquidity. Taken together, the results of the regression analysis are robust, and all point to the conclusion that after the crisis dealers are more reluctant to provide liquidity. Dealers are willing to provide additional immediacy only if they obtain a higher compensation. 7 Conclusion The cost of immediacy for corporate bonds has increased significantly after the 2008 credit crisis. We show that the supply of immediacy has become more elastic with respect to its price. Hence, dealers now charge a higher price for supplying a given amount of immediacy. Traditional dealers have lost market share and dealer behavior 24 has changed after the crisis so that risky bonds are no longer kept on inventory for longer periods of time. All these findings are consistent with the predictions in Duffie (2012) on how the post-crisis regulatory regime would impede market making. This paper adds to the debate about the effects and side-effects of the postcrisis financial regulation. In a study on behalf of SIFMA, Wyman (2012) concluded that the Volcker rule would represent a significant risk for market liquidity. Both Johnson (2012) and Richardson (2012) argued against this conclusion. Although, market liquidity in general has not gone down when looking at traditional liquidity measures, the cost of immediacy has increased as predicted by Duffie (2012). Time constrained traders now pay a higher premium for trading than before the decline in dealer inventories. Whether or not the post-crisis regulation can be considered a success depends on which segments of the economy will be affected by future shocks. By encouraging traditional market makers to take on less risk than before the crisis, the regulation makes a shock less likely to originate in the banking sector (Johnson, 2012; Richardson, 2012). However, the lower willingness to commit capital might make it harder for dealers to mitigate the selling pressure originating from shocks to other segments of the economy. Our research shows that it would be costly to seek liquidity during such fire sales events. 25 References Admanti, Anat R, and Paul Pfleiderer, 1991, Sunshine trading and financial market equilibrium, Review of Financial Studies 4, 443–481. Adrian, Tobias, Erkko Etula, and Tyler Muir, 2014, Financial intermediaries and the crosssection of asset returns, Journal of Finance 69, 2557–2596. Adrian, Tobias, Michael Fleming, Or Shachar, and Erik Vogt, 2015, Has u.s. corporate bond market liquidity deteriorated?, http://libertystreeteconomics.newyorkfed.org/. Ambrose, Brent, Kelly Nianyun Cai, and Jean Helwege, 2012, Fallen angels and price pressure, Journal of Fixed Income 21. Amihud, Yakov, and Haim Mendelson, 1980, Dealership market: Market-making with inventory, Journal of Financial Economics 8, 31–53. Bao, Jack, Maureen O’Hara, and Xing Zhou, 2016, The volcker rule and market-making in times of stress, Working paper Federal Reserve Board. Barber, Brad, and John Lyon, 1997, Detecting long-run abnormal stock returns: The empirical power and specification of test statistics, Journal of Financial Economics 43, 341–372. Beneish, Messod, and Robert Whaley, 1996, An anatomy of the s&p game: The effects of changing the rules, The Journal of Finance 51, 1909–1930. Bessembinder, Hendrik, Allen Carrion, Laura Tuttle, and Kumar Venkataraman, Forthcoming, Liquidity, resiliency and market quality around predictable trades: Theory and evidence, Journal of Financial Economics. Bessembinder, Hendrik, Stacey Jacobsen, William Maxwell, and Kumar Venkataraman, 2016, Capital commitment and illiquidity in corporate bonds, Working paper Arizona State University. Bessembinder, Hendrik, Kathleen Kahle, William Maxwell, and Danielle Xu, 2009, Measuring abnormal bond performance, Review of Financial Studies 22, 4219–4258. Bessembinder, Hendrik, William Maxwell, and Kumar Venkataraman, 2006, Market transparency, liquidity externalities, and institutional trading costs in corporate bonds, Journal of Financial Economics 82, 251–288. 26 Blume, Marshall, and Roger Edelen, 2004, S&p 500 indexers, tracking error, and liquidity, The Journal of Portfolio Management 30, 37–46. Brunnermeier, Markus K, and Lasse Heje Pedersen, 2005, Predatory trading, The Journal of Finance 60, 1825–1863. Cai, Nianyun Kelly, Jean Helwege, and Arthur Warga, 2007, Underpricing in the corporate bond market, Review of Financial Studies 20, 2021–2046. Chacko, George C, Jakub W Jurek, and Erik Stafford, 2008, The price of immediacy, The Journal of Finance 63, 1253–1290. Chen, Honghui, Gregory Noronha, and Vijay Singal, 2004, The price response to s&p 500 index additions and deletions: Evidence of asymmetry and a new explanation, The Journal of Finance 59, 1901–1930. Chen, Zhihua, Aziz Lookman, Norman Schürhoff, and Duane Seppi, 2014, Rating-based investment practices and bond market segmentation, Review of Asset Pricing Studies 4, 162–205. Choi, Darwin, Mila Getmansky, Brian Henderson, and Heather Tookes, 2010, Convertible bond arbitrageurs as suppliers of capital, Review of Financial Studies 23, 2492–2522. Denis, Diane K, John J McConnell, Alexei V Ovtchinnikov, and Yun Yu, 2003, S&p 500 index additions and earnings expectations, The Journal of Finance 58, 1821–1840. Dick-Nielsen, Jens, 2009, Liquidity biases in trace, Journal of Fixed Income 19. , Peter Feldhütter, and David Lando, 2012, Corporate bond liquidity before and after the onset of the subprime crisis, Journal of Financial Economics 103, 471–492. Duffie, Darrell, 2012, Market making under the proposed volcker rule, Rock Center for Corporate Governance at Stanford University Working Paper. Dynkin, Lev, Anthony Gould, and Vadim Konstantinovsky, 2006, Replicating bond indices with liquid derivatives, Journal of Fixed Income 15. Easley, David, and Maureen O’hara, 1987, Price, trade size, and information in securities markets, Journal of Financial Economics 19, 69–90. Edwards, Amy, Lawrence Harris, and Michael Piwowar, 2007, Corporate bond market transaction costs and transparency, The Journal of Finance 62, 1421–1451. 27 Ellul, Andrew, Chotibhak Jotikasthira, and Christian Lundblad, 2011, Regulatory pressure and fire sales in the corporate bond market, Journal of Financial Economics 101, 596– 620. Feldhütter, Peter, 2012, The same bond at different prices: identifying search frictions and selling pressures, Review of Financial Studies 25, 1155–1206. Garman, Mark, 1976, Market microstructure, Journal of Financial Economics 3, 257–275. Goldstein, Michael, and Edith Hotchkiss, 2008, Dealer behavior and the trading of newly issued corporate bonds, Unpublished working paper. Boston College. Greenwood, Robin, 2005, Short- and long-term demand curves for stocks: theory and evidence on the dynamics of arbitrage, Journal of Financial Economics 75, 607–649. Grier, Paul, and Steven Katz, 1976, The differential effects of bond rating changes among industrial and public utility bonds by maturity, Journal of Business pp. 226–239. Harris, Lawrence, and Eitan Gurel, 1986, Price and volume effects associated with changes in the s&p 500 list: New evidence for the existence of price pressures, The Journal of Finance 41, 815–829. Hegde, Shantaram, and John McDermott, 2003, The liquidity effects of revisions to the s&p 500 index: An empirical analysis, Journal of Financial Markets 6, 413–459. Ho, Thomas, and Hans Stoll, 1981, Optimal dealer pricing under transactions and return uncertainty, Journal of Financial Economics 9, 47–73. Johnson, Simon, 2012, Restrictions on proprietary trading and certain interests in and relationships with hedge funds and private equity funds, Comment letter to the SEC. Katz, Steven, 1974, The price and adjustment process of bonds to rating reclassifications: a test of bond market efficiency, Journal of Finance pp. 551–559. Lou, Dong, Hongjun Yan, and Jinfan Zhang, 2013, Anticipated and repeated shocks in liquid markets, Review of Financial Studies 26, 1891–1912. Lucas, Robert, 1976, Econometric policy evaluation: A critique, in Carnegie-Rochester conference series on public policy vol. 1 pp. 19–46. Elsevier. Mase, Bryan, 2007, The impact of changes in the ftse 100 index, Financial Review 42, 461–484. 28 Newman, Yigal, and Michael Rierson, 2004, Illiquidity spillovers: Theory and evidence from european telecom bond issuance, Unpublished working paper. Stanford University. Norden, Lars, and Martin Weber, 2004, Informational efficiency of credit default swap and stock markets: The impact of credit rating announcements, Journal of Banking & Finance 28, 2813–2843. Richardson, Matthew, 2012, Why the volcker rule is a useful tool for managing systemic risk, White paper, New York University. Rossi, Marco, 2014, Realized volatility, liquidity, and corporate yield spreads, The Quarterly Journal of Finance 4, 1450004. Schultz, Paul, 2001, Corporate bond trading costs: A peek behind the curtain, Journal of Finance 56, 677–698. Schwab, 2009, Schwab bond funds prospectus, White Paper. Shleifer, Andrei, 1986, Do demand curves for stocks slope down?, The Journal of Finance 41, 579–590. Stoll, Hans R, 1978, The supply of dealer services in securities markets, The Journal of Finance 33, 1133–1151. Trebbi, Francesco, and Kairong Xiao, 2015, Regulation and market liquidity, Working Paper 21739 National Bureau of Economic Research. Vanguard, 2009, Vanguard bond index funds prospectus, White Paper. Watson, Towers, 2012, Corporate bond liquidity constrained, but active managers can still excel, White Paper. Wyman, Oliver, 2012, The volcker rule restrictions on proprietary trading: Implications for market liquidity, White Paper. 29 Table 1: Barclay Capital corporate bond index exclusion statistics The statistics are accumulated from July 2002 to November 2013 for the Barclay Corporate Bond Index (formerly Lehman). Market value in $1,000 is the average market value at the time of the index revision. The table shows four reasons for being excluded. The maturity of the bonds can fall below 1 year during the month. The bond can be called. The bond can be downgraded from investment-grade to speculative grade during the month. Finally, there can be various other reasons for being excluded. Most of these exclusions are due to revisions of the general index rules, mainly that the size requirement has been increased twice over the period. In all cases the bonds are excluded at the end of the month (last trading day). Reason Maturity< 1 Called Downgrade Other N Market Value ($1,000) OA Duration Coupon 3,102 392 1,078 2,119 645,374 461,354 484,269 358,501 0.92 0.52 5.1 6.0 5.7 7.1 6.8 6.5 30 Table 2: Dealer intertemporal bid-ask spreads: Maturity exclusions This table shows the dealer-bond specific average returns of bonds excluded from the Barclay Corporate Bond Index because of low maturity. Returns are calculated as log price changes between day 0 (the exclusion date) and day t after the exclusion. The returns are calculated as seen from the dealers perspective. First, the intertemporal bid-ask spread is calculated using the dealer-buy price (dealer-specific average buy price over day -2,-1, and 0) and the average dealer sell price at day t (average across all dealers). Second, the abnormal return is the intertemporal bid-ask spread minus the return on a matched portfolio. The portfolio is matched on rating and time to maturity. EW returns are equally weighted across all excluded bonds. VW1 is weighted by the aggregate buying volume in the specific cusip for all dealers with a positive inventory build-up in the bond. VW2 is weighted by the aggregate inventory build-up for dealers with a net positive inventory change between day -3 to 0. The three time periods are 2002Q3-2007Q2, 2007Q3-2009Q4, and 2010Q1-2013Q4. 31 Table 2 (continued) [0, t] N EW Intertemporal Bid-Ask VW1 VW2 EW Abnormal Returns VW1 VW2 22.87 (14.41)∗∗∗ 25.34 (14.71)∗∗∗ 26.63 (15.55)∗∗∗ 29.42 (16.56)∗∗∗ 30.26 (17.19)∗∗∗ 35.49 (18.22)∗∗∗ 51.52 (19.32)∗∗∗ 64.94 (18.71)∗∗∗ 9.34 (9.08)∗∗∗ 12.61 (10.88)∗∗∗ 14.36 (13.04)∗∗∗ 16.05 (13.87)∗∗∗ 17.09 (14.47)∗∗∗ 24.30 (15.65)∗∗∗ 39.69 (10.40)∗∗∗ 54.87 (12.11)∗∗∗ 9.45 (9.90)∗∗∗ 12.34 (10.25)∗∗∗ 14.45 (12.40)∗∗∗ 15.97 (13.24)∗∗∗ 16.79 (12.36)∗∗∗ 23.55 (15.21)∗∗∗ 37.04 (13.04)∗∗∗ 51.80 (12.40)∗∗∗ 20.22 (12.80)∗∗∗ 20.78 (13.06)∗∗∗ 21.15 (12.92)∗∗∗ 23.03 (12.35)∗∗∗ 22.17 (13.12)∗∗∗ 21.29 (12.20)∗∗∗ 22.76 (9.86)∗∗∗ 23.22 (9.88)∗∗∗ 6.34 (9.23)∗∗∗ 7.31 (10.65)∗∗∗ 7.66 (10.05)∗∗∗ 7.87 (7.92)∗∗∗ 7.59 (8.74)∗∗∗ 8.05 (6.22)∗∗∗ 7.20 (8.40)∗∗∗ 7.92 (7.13)∗∗∗ 6.17 (8.04)∗∗∗ 7.13 (8.12)∗∗∗ 7.94 (9.43)∗∗∗ 8.33 (9.41)∗∗∗ 7.74 (7.60)∗∗∗ 8.20 (7.20)∗∗∗ 7.53 (6.82)∗∗∗ 7.50 (6.46)∗∗∗ 55.87 (11.89)∗∗∗ 55.15 (13.82)∗∗∗ 59.07 (10.17)∗∗∗ 63.17 (8.30)∗∗∗ 70.03 (9.65)∗∗∗ 82.52 (10.69)∗∗∗ 122.81 (5.59)∗∗∗ 156.75 (5.20)∗∗∗ 58.36 (6.93)∗∗∗ 57.95 (6.88)∗∗∗ 61.12 (8.14)∗∗∗ 62.12 (7.37)∗∗∗ 67.27 (9.73)∗∗∗ 79.42 (7.89)∗∗∗ 94.50 (5.94)∗∗∗ 130.99 (4.25)∗∗∗ 51.22 (7.49)∗∗∗ 50.15 (6.86)∗∗∗ 58.62 (6.52)∗∗∗ 58.59 (6.54)∗∗∗ 62.56 (7.27)∗∗∗ 73.19 (5.14)∗∗∗ 86.86 (7.08)∗∗∗ 119.01 (4.59)∗∗∗ 46.33 (10.26)∗∗∗ 46.57 (8.12)∗∗∗ 49.80 (7.16)∗∗∗ 52.96 (8.38)∗∗∗ 53.18 (6.23)∗∗∗ 63.28 (7.36)∗∗∗ 76.35 (5.58)∗∗∗ 96.55 (4.66)∗∗∗ 50.43 (6.71)∗∗∗ 50.86 (6.25)∗∗∗ 56.52 (5.70)∗∗∗ 56.89 (7.34)∗∗∗ 56.27 (6.35)∗∗∗ 68.71 (7.00)∗∗∗ 72.47 (4.32)∗∗∗ 102.75 (3.90)∗∗∗ 43.02 (6.50)∗∗∗ 42.12 (5.13)∗∗∗ 52.18 (5.00)∗∗∗ 48.79 (6.35)∗∗∗ 47.12 (6.12)∗∗∗ 54.53 (5.09)∗∗∗ 54.52 (3.11)∗∗∗ 80.71 (3.52)∗∗∗ 27.34 (13.81)∗∗∗ 29.08 (13.99)∗∗∗ 29.04 (12.91)∗∗∗ 32.47 (12.16)∗∗∗ 33.71 (12.06)∗∗∗ 36.16 (12.14)∗∗∗ 44.38 (9.06)∗∗∗ 51.82 (8.71)∗∗∗ 14.15 (9.27)∗∗∗ 15.21 (11.45)∗∗∗ 15.38 (11.90)∗∗∗ 16.57 (9.36)∗∗∗ 17.38 (8.82)∗∗∗ 19.36 (9.01)∗∗∗ 22.90 (6.87)∗∗∗ 27.64 (7.19)∗∗∗ 13.67 (8.96)∗∗∗ 14.64 (10.76)∗∗∗ 14.89 (11.82)∗∗∗ 15.84 (8.98)∗∗∗ 16.85 (8.63)∗∗∗ 18.70 (8.75)∗∗∗ 21.85 (6.78)∗∗∗ 25.31 (7.22)∗∗∗ 26.27 (12.76)∗∗∗ 27.16 (13.70)∗∗∗ 26.47 (12.83)∗∗∗ 29.46 (12.22)∗∗∗ 30.06 (12.29)∗∗∗ 30.19 (13.38)∗∗∗ 34.06 (10.49)∗∗∗ 34.20 (10.39)∗∗∗ 13.53 (8.24)∗∗∗ 13.79 (9.94)∗∗∗ 13.25 (10.15)∗∗∗ 13.99 (8.64)∗∗∗ 14.35 (7.80)∗∗∗ 14.87 (9.24)∗∗∗ 15.93 (9.55)∗∗∗ 15.09 (9.42)∗∗∗ 13.30 (8.54)∗∗∗ 13.59 (10.12)∗∗∗ 13.06 (10.15)∗∗∗ 13.62 (8.73)∗∗∗ 14.08 (7.84)∗∗∗ 14.46 (9.23)∗∗∗ 16.02 (9.23)∗∗∗ 14.37 (8.73)∗∗∗ Period 1 1 830 2 794 3 780 4 777 5 763 10 727 20 688 30 675 Period 2 1 269 2 254 3 236 4 235 5 230 10 211 20 211 30 206 Period 3 1 1,085 2 1,054 3 1,041 4 995 5 990 10 954 20 861 30 814 32 Table 3: Dealer intertemporal bid-ask spreads: Downgrade exclusions This table shows the dealer-bond specific returns of bond excluded from the Barclay Corporate Bond Index because of downgrade from investment-grade to speculative grade. Returns are calculated as log price changes between day 0 (the exclusion date) and day t after the exclusion. The returns are calculated as seen from the dealers perspective. First, the intertemporal bid-ask spread is calculated using the dealer-buy price (dealer-specific average buy price over day -2,-1, and 0) and the average dealer sell price at day t (average across all dealers). Second, the abnormal return is the intertemporal bid-ask spread minus the return on a matched portfolio. The portfolio is matched on rating and time to maturity. EW returns are equally weighted across all excluded bonds. VW1 is weighted by the aggregate buying volume in the specific cusip for all dealers with a positive inventory build-up in the bond. VW2 is weighted by the aggregate inventory build-up for dealers with a net positive inventory change between day -3 to 0. The three time periods are 2002Q3-2007Q2, 2007Q3-2009Q4, and 2010Q1-2013Q4. 33 Table 3 (continued) [0, t] N EW Intertemporal Bid-Ask VW1 VW2 EW Abnormal Returns VW1 VW2 139.73 (4.62)∗∗∗ 250.70 (3.41)∗∗∗ 282.37 (3.67)∗∗∗ 267.30 (4.95)∗∗∗ 318.45 (5.04)∗∗∗ 389.65 (3.60)∗∗∗ 367.99 (5.06)∗∗∗ 422.03 (3.16)∗∗∗ 162.06 (10.63)∗∗∗ 333.73 (8.14)∗∗∗ 368.52 (9.53)∗∗∗ 328.41 (16.59)∗∗∗ 389.00 (19.10)∗∗∗ 500.46 (10.25)∗∗∗ 399.81 (24.79)∗∗∗ 513.42 (12.15)∗∗∗ 147.54 (7.64)∗∗∗ 307.93 (6.25)∗∗∗ 335.56 (7.27)∗∗∗ 308.38 (13.56)∗∗∗ 367.67 (15.73)∗∗∗ 460.66 (7.51)∗∗∗ 373.50 (19.97)∗∗∗ 470.94 (9.80)∗∗∗ 98.19 (4.39)∗∗∗ 157.89 (3.81)∗∗∗ 160.68 (3.96)∗∗∗ 168.71 (4.96)∗∗∗ 193.42 (5.21)∗∗∗ 251.28 (3.50)∗∗∗ 173.15 (4.70)∗∗∗ 173.25 (2.72)∗∗∗ 96.18 (8.15)∗∗∗ 188.18 (7.21)∗∗∗ 184.35 (7.60)∗∗∗ 195.90 (9.86)∗∗∗ 220.81 (10.94)∗∗∗ 295.58 (8.20)∗∗∗ 154.53 (9.08)∗∗∗ 174.94 (8.46)∗∗∗ 81.19 (5.73)∗∗∗ 166.18 (5.10)∗∗∗ 155.62 (5.12)∗∗∗ 172.19 (7.06)∗∗∗ 196.05 (7.92)∗∗∗ 256.95 (5.29)∗∗∗ 124.79 (6.40)∗∗∗ 142.36 (6.76)∗∗∗ 48.13 (0.75) 70.68 (0.62) 14.65 (0.09) 30.42 (0.14) 65.02 (0.25) 241.79 (1.01) 442.86 (1.86)∗ 125.36 (0.31) 65.43 (1.17) 76.11 (0.92) 87.71 (0.79) 155.40 (0.89) 193.97 (0.89) 296.57 (1.69)∗ 467.30 (2.97)∗∗∗ 412.56 (3.79)∗∗∗ 110.55 (1.78)∗ 135.18 (1.77)∗ 161.95 (4.96)∗∗∗ 233.01 (2.19)∗∗ 334.67 (2.47)∗∗ 463.31 (2.85)∗∗∗ 576.55 (2.35)∗∗ 427.40 (4.49)∗∗∗ 58.60 (1.36) 80.74 (1.08) 42.16 (0.47) 50.22 (0.37) 82.48 (0.53) 162.46 (0.93) 234.47 (1.15) -139.2 (-0.37) 59.14 (1.59) 65.61 (1.26) 74.38 (1.08) 118.90 (0.96) 152.41 (0.98) 193.45 (1.33) 334.32 (1.97)∗∗ 270.88 (1.65)∗ 93.56 (1.67)∗ 112.61 (1.88)∗ 130.23 (7.26)∗∗∗ 174.51 (2.12)∗∗ 260.06 (2.46)∗∗ 344.43 (2.83)∗∗∗ 492.31 (2.84)∗∗∗ 373.44 (3.16)∗∗∗ 123.91 (1.19) 193.05 (1.70)∗ 240.29 (1.66)∗ 263.72 (1.55) 301.42 (1.59) 199.79 (1.74)∗ 398.41 (1.54) 538.57 (1.70)∗ 365.93 (2.61)∗∗∗ 474.34 (2.69)∗∗∗ 644.88 (2.71)∗∗∗ 757.23 (2.69)∗∗∗ 844.22 (2.87)∗∗∗ 413.16 (2.50)∗∗ 1061.0 (2.77)∗∗∗ 1407.1 (3.19)∗∗∗ 372.42 (2.85)∗∗∗ 494.84 (3.06)∗∗∗ 667.15 (2.93)∗∗∗ 770.70 (2.85)∗∗∗ 874.81 (3.04)∗∗∗ 475.31 (2.59)∗∗∗ 1136.2 (3.24)∗∗∗ 1434.4 (3.33)∗∗∗ 99.91 (1.14) 149.92 (1.76)∗ 185.00 (1.69)∗ 203.06 (1.58) 231.84 (1.59) 173.39 (1.71)∗ 314.30 (1.62) 332.27 (1.45) 292.93 (2.65)∗∗∗ 350.12 (2.79)∗∗∗ 488.51 (2.80)∗∗∗ 577.79 (2.83)∗∗∗ 651.57 (2.98)∗∗∗ 381.56 (2.60)∗∗∗ 807.29 (2.87)∗∗∗ 937.37 (2.99)∗∗∗ 294.79 (2.84)∗∗∗ 366.33 (3.16)∗∗∗ 508.88 (3.04)∗∗∗ 592.19 (3.07)∗∗∗ 682.85 (3.21)∗∗∗ 444.73 (2.83)∗∗∗ 869.89 (3.37)∗∗∗ 965.48 (3.11)∗∗∗ Period 1 1 243 2 245 3 243 4 234 5 229 10 226 20 215 30 209 Period 2 1 107 2 101 3 102 4 93 5 87 10 91 20 77 30 71 Period 3 1 213 2 208 3 193 4 185 5 188 10 177 20 175 30 163 34 Table 4: Aggregate intertemporal bid-ask spreads: maturity exclusions This table replicates Table 2, except that the time zero price is an average (across all dealers) of the transaction prices at time zero. [0, t] N EW Intertemporal Bid-Ask VW1 VW2 EW Abnormal Returns VW1 VW2 7.58 (10.26)∗∗∗ 9.25 (9.39)∗∗∗ 10.80 (8.78)∗∗∗ 13.14 (10.23)∗∗∗ 14.81 (11.80)∗∗∗ 19.78 (13.14)∗∗∗ 31.43 (13.32)∗∗∗ 46.04 (14.65)∗∗∗ 5.81 (8.65)∗∗∗ 8.19 (8.81)∗∗∗ 9.50 (10.09)∗∗∗ 11.12 (12.34)∗∗∗ 12.43 (12.74)∗∗∗ 18.81 (15.61)∗∗∗ 32.74 (11.95)∗∗∗ 45.83 (12.27)∗∗∗ 5.96 (8.88)∗∗∗ 8.10 (7.99)∗∗∗ 9.84 (9.74)∗∗∗ 11.46 (13.01)∗∗∗ 12.38 (11.77)∗∗∗ 18.50 (13.09)∗∗∗ 31.28 (13.30)∗∗∗ 44.14 (12.20)∗∗∗ 5.41 (7.85)∗∗∗ 5.87 (8.36)∗∗∗ 6.85 (7.16)∗∗∗ 8.47 (8.11)∗∗∗ 9.08 (8.98)∗∗∗ 10.57 (10.36)∗∗∗ 15.33 (11.65)∗∗∗ 23.48 (13.33)∗∗∗ 3.86 (8.41)∗∗∗ 5.01 (7.43)∗∗∗ 5.56 (7.48)∗∗∗ 6.60 (9.94)∗∗∗ 6.99 (9.21)∗∗∗ 9.25 (10.59)∗∗∗ 14.84 (11.41)∗∗∗ 21.36 (11.11)∗∗∗ 4.02 (8.51)∗∗∗ 5.10 (6.66)∗∗∗ 6.06 (7.44)∗∗∗ 7.23 (12.09)∗∗∗ 7.21 (8.76)∗∗∗ 9.48 (9.56)∗∗∗ 14.40 (11.76)∗∗∗ 20.71 (11.02)∗∗∗ 56.47 (9.10)∗∗∗ 64.08 (7.58)∗∗∗ 65.85 (8.06)∗∗∗ 68.50 (8.32)∗∗∗ 73.45 (8.66)∗∗∗ 81.24 (8.10)∗∗∗ 107.91 (7.45)∗∗∗ 143.19 (5.71)∗∗∗ 46.28 (8.16)∗∗∗ 48.05 (6.33)∗∗∗ 54.17 (7.78)∗∗∗ 53.68 (6.49)∗∗∗ 62.21 (8.32)∗∗∗ 70.81 (8.84)∗∗∗ 92.59 (5.47)∗∗∗ 122.32 (4.21)∗∗∗ 44.04 (7.76)∗∗∗ 46.30 (6.66)∗∗∗ 51.74 (7.62)∗∗∗ 56.07 (7.80)∗∗∗ 63.55 (7.73)∗∗∗ 74.47 (6.18)∗∗∗ 88.18 (5.10)∗∗∗ 121.24 (4.07)∗∗∗ 48.45 (8.47)∗∗∗ 54.24 (7.73)∗∗∗ 55.56 (7.00)∗∗∗ 55.91 (7.50)∗∗∗ 58.45 (7.33)∗∗∗ 68.48 (8.62)∗∗∗ 76.90 (9.09)∗∗∗ 99.60 (7.92)∗∗∗ 38.19 (8.50)∗∗∗ 39.72 (6.10)∗∗∗ 46.97 (5.35)∗∗∗ 46.90 (6.66)∗∗∗ 49.97 (5.33)∗∗∗ 65.73 (8.94)∗∗∗ 77.35 (6.51)∗∗∗ 98.44 (6.69)∗∗∗ 37.53 (6.80)∗∗∗ 39.06 (5.86)∗∗∗ 44.35 (6.26)∗∗∗ 49.08 (7.75)∗∗∗ 51.16 (6.90)∗∗∗ 66.49 (6.89)∗∗∗ 72.14 (7.65)∗∗∗ 95.83 (7.02)∗∗∗ 13.72 (10.49)∗∗∗ 14.93 (10.96)∗∗∗ 15.15 (10.75)∗∗∗ 16.72 (9.78)∗∗∗ 17.11 (10.30)∗∗∗ 19.04 (9.72)∗∗∗ 24.08 (7.96)∗∗∗ 29.09 (7.67)∗∗∗ 11.29 (9.68)∗∗∗ 12.83 (10.65)∗∗∗ 12.74 (10.99)∗∗∗ 13.73 (9.68)∗∗∗ 14.74 (8.24)∗∗∗ 16.80 (8.15)∗∗∗ 20.64 (5.95)∗∗∗ 23.49 (6.29)∗∗∗ 13.18 (8.21)∗∗∗ 14.29 (8.75)∗∗∗ 14.88 (6.38)∗∗∗ 16.28 (6.86)∗∗∗ 17.39 (6.76)∗∗∗ 20.13 (5.86)∗∗∗ 24.60 (4.54)∗∗∗ 34.06 (3.25)∗∗∗ 12.68 (10.42)∗∗∗ 13.46 (10.83)∗∗∗ 13.26 (10.84)∗∗∗ 14.63 (9.93)∗∗∗ 14.62 (10.57)∗∗∗ 15.03 (10.53)∗∗∗ 18.17 (9.13)∗∗∗ 19.69 (8.06)∗∗∗ 10.47 (9.57)∗∗∗ 11.54 (9.70)∗∗∗ 11.05 (10.29)∗∗∗ 11.82 (9.81)∗∗∗ 12.53 (7.54)∗∗∗ 13.23 (7.79)∗∗∗ 15.82 (6.08)∗∗∗ 15.21 (6.10)∗∗∗ 12.49 (8.25)∗∗∗ 13.14 (8.27)∗∗∗ 13.26 (6.20)∗∗∗ 14.49 (6.73)∗∗∗ 15.40 (6.42)∗∗∗ 16.95 (5.39)∗∗∗ 21.08 (3.86)∗∗∗ 27.16 (2.54)∗∗ Period 1 1 834 2 798 3 782 4 780 5 766 10 731 20 690 30 683 Period 2 1 274 2 258 3 237 4 239 5 234 10 218 20 216 30 213 Period 3 1 1,072 2 1,042 3 1,030 4 984 5 981 10 942 20 853 30 804 35 Table 5: Aggregate intertemporal bid-ask spreads: maturity exclusions This table replicates Table 3, except that the time zero price is an average (across all dealers) of the transaction prices at time zero. [0, t] N EW Intertemporal Bid-Ask VW1 VW2 EW Abnormal Returns VW1 VW2 100.74 (4.77)∗∗∗ 149.11 (3.03)∗∗∗ 168.44 (3.30)∗∗∗ 177.16 (4.28)∗∗∗ 203.93 (3.99)∗∗∗ 232.72 (2.94)∗∗∗ 255.84 (4.66)∗∗∗ 277.82 (2.62)∗∗∗ 107.76 (4.80)∗∗∗ 207.20 (3.20)∗∗∗ 256.85 (4.18)∗∗∗ 242.64 (5.86)∗∗∗ 278.99 (4.87)∗∗∗ 371.70 (4.51)∗∗∗ 319.06 (4.61)∗∗∗ 322.58 (2.13)∗∗ 91.70 (6.37)∗∗∗ 181.77 (3.20)∗∗∗ 215.31 (3.87)∗∗∗ 216.32 (5.49)∗∗∗ 259.63 (5.60)∗∗∗ 295.16 (4.08)∗∗∗ 274.72 (5.62)∗∗∗ 280.44 (2.88)∗∗∗ 78.25 (5.19)∗∗∗ 108.60 (3.37)∗∗∗ 115.43 (3.83)∗∗∗ 127.83 (4.68)∗∗∗ 147.95 (4.61)∗∗∗ 168.31 (3.04)∗∗∗ 156.32 (4.98)∗∗∗ 161.02 (2.46)∗∗ 81.91 (3.64)∗∗∗ 153.80 (4.29)∗∗∗ 177.85 (5.62)∗∗∗ 182.74 (7.45)∗∗∗ 200.58 (6.31)∗∗∗ 283.79 (5.39)∗∗∗ 205.81 (6.16)∗∗∗ 187.28 (2.22)∗∗ 63.95 (7.23)∗∗∗ 125.45 (3.83)∗∗∗ 137.42 (5.20)∗∗∗ 152.86 (6.90)∗∗∗ 176.04 (7.06)∗∗∗ 205.26 (4.89)∗∗∗ 159.11 (6.24)∗∗∗ 142.17 (2.72)∗∗∗ 244.81 (2.34)∗∗ 248.08 (1.63) 219.47 (1.39) 61.79 (0.56) 211.29 (1.35) -2.59 (-0.01) 502.34 (1.47) 482.59 (1.34) 260.68 (2.90)∗∗∗ 274.92 (1.59) 219.60 (0.93) 64.76 (0.55) 259.21 (1.88)∗ -72.72 (-0.29) 258.53 (0.76) 366.42 (2.17)∗∗ 251.98 (4.14)∗∗∗ 327.58 (2.63)∗∗∗ 315.24 (2.43)∗∗ 195.05 (2.65)∗∗∗ 363.23 (3.88)∗∗∗ 3.35 (0.01) 231.89 (0.85) 425.21 (2.99)∗∗∗ 276.21 (2.23)∗∗ 278.59 (1.59) 264.11 (1.51) 228.01 (1.23) 389.21 (1.66)∗ 484.20 (2.04)∗∗ 731.14 (2.17)∗∗ 653.52 (1.51) 337.68 (2.68)∗∗∗ 365.23 (1.71)∗ 307.07 (1.20) 417.62 (2.07)∗∗ 603.17 (2.55)∗∗ 857.76 (2.13)∗∗ 840.43 (3.44)∗∗∗ 939.52 (2.04)∗∗ 338.27 (3.33)∗∗∗ 448.19 (2.38)∗∗ 442.94 (2.24)∗∗ 602.66 (2.60)∗∗∗ 781.29 (2.86)∗∗∗ 1076.9 (2.84)∗∗∗ 910.67 (3.54)∗∗∗ 1129.3 (2.32)∗∗ 146.26 (2.25)∗∗ 230.77 (2.59)∗∗∗ 314.24 (2.29)∗∗ 369.22 (2.15)∗∗ 414.73 (2.32)∗∗ 361.26 (2.35)∗∗ 497.84 (2.14)∗∗ 690.20 (2.44)∗∗ 144.95 (1.88)∗ 272.94 (2.72)∗∗∗ 357.19 (2.50)∗∗ 396.80 (2.25)∗∗ 417.02 (2.09)∗∗ 278.32 (1.74)∗ 489.02 (1.97)∗∗ 715.33 (2.25)∗∗ 133.39 (2.34)∗∗ 224.79 (2.94)∗∗∗ 307.21 (2.66)∗∗∗ 352.28 (2.44)∗∗ 388.10 (2.47)∗∗ 303.73 (2.64)∗∗∗ 497.66 (2.24)∗∗ 736.24 (2.51)∗∗ 115.29 (2.43)∗∗ 182.03 (2.90)∗∗∗ 251.97 (2.45)∗∗ 296.71 (2.27)∗∗ 336.39 (2.46)∗∗ 333.66 (2.28)∗∗ 391.35 (2.06)∗∗ 494.07 (2.26)∗∗ 120.07 (1.99)∗∗ 227.37 (3.06)∗∗∗ 296.31 (2.76)∗∗∗ 327.72 (2.52)∗∗ 339.23 (2.23)∗∗ 264.38 (1.74)∗ 401.71 (2.02)∗∗ 509.90 (2.09)∗∗ 104.38 (2.47)∗∗ 175.36 (3.36)∗∗∗ 247.27 (2.94)∗∗∗ 286.97 (2.76)∗∗∗ 313.15 (2.75)∗∗∗ 283.66 (2.76)∗∗∗ 407.29 (2.37)∗∗ 534.25 (2.44)∗∗ Period 1 1 244 2 246 3 244 4 235 5 229 10 227 20 215 30 210 Period 2 1 147 2 145 3 138 4 126 5 119 10 124 20 104 30 97 Period 3 1 211 2 207 3 191 4 183 5 185 10 179 20 176 30 162 36 Table 6: Market share for the most active dealers The table shows the market share for the most active dealers. The market share is calculated as the fraction of total event dealer-buying volume that the most active dealers facilitated. The most active dealers are defined as the 4 dealers that on average facilitated the most event buying volume and which were alive in the entire sample period. The periods are 2002Q3-2007Q2 (pre-crisis), 2007Q3-2009Q4 (crisis), and 2010Q1-2013Q4 (post-crisis). The numbers reported are the average combined market share for the most active dealers. Maturity exclusion Downgrade exclusion Pre-Crisis Crisis Post-Crisis 0.212 0.320 0.108 0.183 0.128 0.235 37 Table 7: Descriptive statistics of regression variables This table presents descriptive statistics for the variables used in the regression analysis. The statistics is divided into the whole sample, the downgrade sample, and the low maturity sample. TED spread is the difference between the 3-month LIBOR rate and the 3-month T-bill rate. VIX is the CBOR volatility index derived from the implied volatility on S&P 500 index options. Coupon is the nominal annualized coupon on the corporate bond. Issue size is the offering amount for the bond in millions. Years to maturity is the remaining time to maturity of the bond. Dealer Lev Growth is the aggregate leverage growth for broker-dealers obtained from the Federal Reserve Flow of Funds data. Q is the dealer-specific aggregate imbalance. Rating is an index indicating all rating notches that a bond can take, and takes all integer values from 1 (AAA) to 21 (C). Financials is a dummy indicating whether the bond was issued by a financial institution (according to FISD). Variable Obs. Mean St. Dev. p1 p25 p50 p75 p99 16.50 14.60 -0.03 4.88 6.00 300.00 0.92 17.24 0.00 0.00 0.00 0.00 0.00 23.00 17.92 0.02 5.88 7.00 500.00 0.95 17.71 0.00 0.01 0.03 0.00 0.01 39.50 24.57 0.04 6.95 10.00 750.00 0.99 18.18 1.00 2.60 2.88 0.83 2.86 312.00 55.28 0.65 9.30 13.00 3000.0 27.37 19.55 1.00 89.99 118.83 68.33 82.26 4.63 5.00 300.00 0.92 17.25 0.00 0.00 0.00 0.00 0.00 5.65 6.00 500.00 0.95 17.74 0.00 0.03 0.08 0.00 0.02 6.75 8.00 750.00 0.96 18.19 1.00 3.06 3.76 0.41 4.22 8.95 10.00 3000.0 1.00 19.56 1.00 73.00 60.37 28.22 82.74 5.70 11.00 300.00 3.66 17.07 0.00 0.00 0.00 0.00 0.00 6.63 11.00 500.00 6.59 17.46 0.00 0.00 0.00 0.01 0.00 7.50 12.00 750.00 10.92 18.02 1.00 1.70 2.11 5.00 0.80 9.98 15.00 3000.0 41.44 19.49 1.00 128.37 205.38 100.36 82.26 Panel A: all exclusions TED Spread VIX Dealer Lev. Growth Coupon Rating (1=AAA - 21=C) Issue Size (MIO) Years to Maturity Log Issue Size Financials Dealer Inventory (Q) Pre-crisis Q Crisis Q Post-crisis Q 136 136 136 3,314 3,314 3,314 3,314 3,314 3,314 223,603 86,592 24,829 112,182 44.10 20.82 -0.01 5.80 7.52 664.91 2.84 17.75 0.43 6.51 7.89 2.85 6.25 Coupon Rating (1=AAA - 21=C) Issue Size (MIO) Years to Maturity Log Issue Size Financials Dealer Inventory (Q) Pre-crisis Q Crisis Q Post-crisis Q 2,619 2,619 2,619 2,619 2,619 2,619 153,304 49,902 19,630 83,772 5.58 6.47 667.38 0.94 17.80 0.46 5.40 4.88 1.44 6.63 Coupon Rating (1=AAA - 21=C) Issue Size (MIO) Years to Maturity Log Issue Size Financials Dealer Inventory (Q) Pre-crisis Q Crisis Q Post-crisis Q 695 695 695 695 695 695 70,299 36,690 5,199 28,410 59.70 9.05 0.20 1.66 2.85 569.02 6.03 0.68 0.50 24.22 32.61 11.62 17.73 6.00 10.91 -0.98 1.38 1.00 200.00 0.87 16.56 0.00 -11.52 -7.73 -9.87 -16.18 Panel B: maturity exclusions 1.67 2.19 562.35 0.03 0.65 0.50 15.45 12.33 6.40 18.20 1.25 1.00 200.00 0.86 16.84 0.00 -13.23 -6.59 -9.84 -19.23 Panel C: downgrade exclusions 6.61 11.46 655.59 10.01 17.57 0.31 8.92 11.98 8.21 5.10 1.30 0.93 593.82 10.42 0.75 0.46 36.57 47.68 21.30 16.23 38 3.50 11.00 150.00 1.11 16.14 0.00 -8.46 -8.51 -10.08 -8.28 39 Number of Observations Adjusted R-Square TED Spread VIX Dealer Lev. Growth Log Issue Size Q*Precrisis Q*Crisis Q*Postcrisis Event Window: (0,t] 15713 0.3233 0.55 (1.77)∗ 1.04 (1.29) 0.11 (1.74)∗ -17.81 (-1.49) -92.15 (-1.97)∗∗ 3.07 (2.27)∗∗ 1.26 (3.19)∗∗∗ 1 15338 0.2935 0.72 (1.74)∗ 1.87 (1.67)∗ 0.25 (1.35) -21.55 (-1.47) -32.71 (-0.71) 2.81 (1.89)∗ 1.36 (3.00)∗∗∗ 2 14993 0.3407 1.00 (2.61)∗∗∗ 2.39 (2.19)∗∗ 0.19 (1.14) -22.75 (-1.39) -77.85 (-1.28) 3.83 (2.08)∗∗ 1.93 (3.35)∗∗∗ 3 14779 0.3244 1.19 (2.46)∗∗ 4.51 (3.01)∗∗∗ 0.13 (0.85) -6.29 (-0.29) -10.90 (-0.11) 2.74 (1.25) 2.37 (3.12)∗∗∗ 4 14634 0.3480 1.41 (2.90)∗∗∗ 5.39 (2.84)∗∗∗ 0.17 (0.78) -16.81 (-0.66) -96.80 (-0.82) 2.81 (1.19) 2.38 (2.85)∗∗∗ 5 14101 0.3404 0.69 (2.77)∗∗∗ 4.33 (2.17)∗∗ 0.11 (0.32) 14.93 (0.73) -35.89 (-0.40) 3.01 (1.75)∗ 1.40 (2.08)∗∗ 10 13401 0.4126 1.95 (2.47)∗∗ 6.12 (2.54)∗∗ -0.03 (-0.09) 2.47 (0.09) -189.4 (-1.33) 7.91 (2.38)∗∗ 1.83 (1.67)∗ 20 12919 0.3287 2.17 (2.03)∗∗ 5.58 (2.33)∗∗ 0.19 (0.45) 68.36 (1.25) -220.1 (-0.87) 5.42 (1.05) 3.11 (1.50) 30 This table presents regression coefficients for a series of regressions. The dependent variable is the bond- and dealer-specific abnormal return over the period from the exclusion day 0 to day t after the exclusion. Variables are defined in Tabe 7. The regressions include (interacted) period/industry/rating and dealer fixed effects. The periods are 2002Q3-2007Q2 (pre-crisis), 2007Q3-2009Q4 (crisis), and 2010Q1-2013Q4 (post-crisis). We consider two industry categories (financial Vs non-financial) and six rating categories (AAA-AA, A, BBB, BB, B, CCC-C). Robust standard errors are clustered at the bond issuer level. Table 8: Liquidity provision before and after the crisis 40 Number of Observations Adjusted R-Square TED Spread VIX Dealer Lev. Growth Log Issue Size Q*Precrisis Q*Crisis Q*Postcrisis Event Window: (0,t] 15713 0.3215 11.44 (1.73)∗ 12.83 (1.34) 3.65 (0.92) -16.74 (-1.41) -89.39 (-1.88)∗ 3.19 (2.27)∗∗ 1.25 (3.19)∗∗∗ 1 15338 0.2918 14.50 (1.74)∗ 32.80 (2.52)∗∗ 12.53 (1.81)∗ -20.46 (-1.44) -34.90 (-0.74) 2.96 (1.91)∗ 1.35 (2.94)∗∗∗ 2 14993 0.3427 20.69 (2.65)∗∗∗ 53.25 (3.32)∗∗∗ 11.86 (1.55) -21.07 (-1.31) -88.22 (-1.45) 4.11 (2.17)∗∗ 1.90 (3.32)∗∗∗ 3 14779 0.3191 23.01 (2.22)∗∗ 62.72 (2.96)∗∗∗ 9.04 (1.47) -3.41 (-0.16) -4.14 (-0.04) 2.97 (1.32) 2.27 (3.03)∗∗∗ 4 14634 0.3423 30.39 (2.66)∗∗∗ 71.27 (2.87)∗∗∗ 14.51 (1.75)∗ -13.63 (-0.54) -84.93 (-0.69) 3.06 (1.26) 2.28 (2.76)∗∗∗ 5 14101 0.3347 18.53 (2.91)∗∗∗ 49.64 (2.96)∗∗∗ 11.46 (0.81) 16.32 (0.78) -18.62 (-0.19) 3.02 (1.70)∗ 1.27 (1.91)∗ 10 This table replicates Table 8, except that the natural logarithm of dealer inventory is used to proxy for Q. Table 9: Liquidity provision (Q) measured in natural logs. 13401 0.4056 40.54 (2.36)∗∗ 66.80 (2.25)∗∗ 5.67 (0.42) 6.23 (0.23) -156.7 (-1.04) 8.21 (2.38)∗∗ 1.74 (1.59) 20 12919 0.3281 49.46 (2.06)∗∗ 83.76 (2.24)∗∗ 14.62 (0.86) 73.08 (1.34) -197.0 (-0.76) 5.80 (1.10) 3.10 (1.51) 30 Table 10: Bond purchase price and the provision of immediacy This table is similar to Table 8, except that the dependent variable is given by bond prices rather than abnormal returns. The specification incudes two additional bond characteristics (coupon and time to maturity) to better control for factors affecting bond prices (but not necessarily returns). Model Q*Postcrisis Q*Crisis Q*Precrisis Log Issue Size Coupon Years to Maturity 1 2 -0.07 (-4.04)∗∗∗ 0.12 (2.28)∗∗ -0.01 (-1.35) 2.97 (4.70)∗∗∗ 1.52 (7.00)∗∗∗ -0.42 (-3.98)∗∗∗ -0.05 (-4.04)∗∗∗ 0.02 (0.41) -0.01 (-1.27) 2.63 (5.04)∗∗∗ 1.59 (7.79)∗∗∗ -0.44 (-4.72)∗∗∗ 11.87 (4.70)∗∗∗ -0.28 (-4.92)∗∗∗ -0.07 (-4.61)∗∗∗ 17415 0.6381 17415 0.7125 Dealer Lev. Growth VIX TED Spread Number of Observations Adjusted R-Square 41 3 2 1 0 0 10 −1 Market Illiqudity Factor 4 5 30 25 20 15 Corp. Bond Inventory (USD bn) 300 250 200 150 100 50 Corp. Sec. Inventory (USD bn) Corporate Securities Corporate bonds Jan11 Jan15 Jan03 (a) Primary dealer inventory Jan11 Jan15 (b) Market Illiquidity Factor 20 1000 22 1200 24 26 Turnover New Issuances 16 800 18 Daily Turnover (USD bn) 7000 6000 5000 14 4000 Market outstanding (USD bn) Jan07 1400 Jan07 2002 2005 2008 2011 2014 2002 (c) Market Size 2005 2008 2011 2014 (d) Market Turnover Figure 1: Corporate bond market statistics Panel A shows the primary dealer inventories in corporate securities (investment-grade above 1 year in maturity) and in corporate bonds. The first series can be retrieved from the New York Fed statistics on primary dealer holdings. The graph on corporate bonds can be retrieved from the same place after March 2003. The numbers prior to that date have been computed by Goldman Sachs using yearly SEC-filings from the primary dealers. Panel B shows the market liquidity measure from Dick-Nielsen, Feldhütter, and Lando (2012) which can be downloaded from peterfeldhutter.com. Panel C shows total nominal corporate bond market size. Panel D shows total market trading volume and the number of total size of new issuances. Data for panel C and D are retrieved from SIFMA. 42 New Issuances (USD bn) Jan03 200 150 100 50 Average daily volume (USD millions) −100 −50 0 50 100 50 100 Event Day 200 150 100 50 Average daily volume (USD millions) (a) Maturity < 1 Year −100 −50 0 Event Day (b) Rating Less Then investment-grade Figure 2: Trading activity around the event. This graphs show the average trading volume around the monthly exclusions. Panel A shows the trading volume for the bonds excluded because of low maturity. Panel B is for the bonds excluded because of a downgrade to speculative grade. Trading volume is aggregated across all the bonds excluded at a given event date and then averaged across all event dates. 43 (a) Maturity < 1 Year (b) Rating Less Then investment-grade Figure 3: Index Exclusions Over time This figure plots the number of bond (square) and firm (circle) exclusions from the Barclay’s investment-grade Index. The top panel presents the exclusions due to maturity; the bottom panel presents the exclusions due to rating deterioration. The shaded area represents the sub-prime crisis. 44 150 100 50 0 Cum. dealer inventory (USD millions) −100 −50 0 50 100 50 100 Event Day 100 80 60 40 20 0 Cum. dealer inventory (USD millions) (a) Maturity < 1 Year −100 −50 0 Event Day (b) Rating Less Then investment-grade Figure 4: Cumulative dealer inventory around the event date. This graphs show the average cumulative dealer inventory around the monthly exclusions. Panel A shows the inventory for the bonds excluded because of low maturity. Panel B is for the bonds excluded because of a downgrade to speculative grade. Cumulative inventory is found by subtracting dealer sells from dealer buys and cumulating the imbalance over time. The dealer inventory is relative to the arbitrarily chosen starting point at event day -100. Inventory is aggregated across all the bonds excluded at a given date and then averaged across all the event dates. 45 50 100 150 200 250 0 −100 Cum. dealer inventory (USD millions) Pre−Crisis Crisis Post−Crisis −20 0 20 40 60 80 100 Event Day 50 100 Pre−Crisis Crisis Post−Crisis 0 Cum. dealer inventory (USD millions) 150 (a) Maturity < 1 Year −20 0 20 40 60 80 100 Event Day (b) Rating Less Then investment-grade Figure 5: Cumulative dealer inventory around the event date for subperiods. This graphs shows the cumulative dealer inventories for three periods. Pre-crisis: 2002Q2 to 2007Q2, Crisis: 2007Q3 to 2009Q4, and Post-crisis: 2010Q1 to 2013Q4. The cumulative inventory and the two panels are calculated as in Figure 4, except that the referencing point is now event day -30. 46 (a) Maturity (EW) (b) Downgrade (EW) (c) Maturity (VW1) (d) Downgrade (VW1) (e) Maturity (VW2) (f) Downgrade (VW2) Figure 6: Cost of immediacy before and after the credit crisis (dealer level) The figure provides a graphical representation of the estimates from the last three columns (abnormal returns) in Table 2 (left graphs) and Table 3 (right graphs). 47 250 200 150 100 50 0 Average daily volume (USD millions) −100 −50 0 50 100 Event Day 40 20 0 −20 Cum. dealer inventory (USD millions) 60 (a) Volume at downgrade date −50 0 50 100 Event Day (b) Cumulative inventory at downgrade date Figure 7: Trading and inventory around the downgrade date. This graphs show the average trading volume and cumulative dealer inventory around the downgrade date. The downgrade date is the date at which the bond changes index rating from investmentgrade to speculative-grade. Trading volume is aggregated across all the downgraded bonds. The cumulative inventory is calculated as in Figure 4, except that the referencing point is now event day -50 and event time is now relative to the downgrade date. 48 −10 0 10 1500 1000 500 0 Cum. dealer inventory (USD millions) 2000 4000 3000 2000 1000 Total Daily volume (USD millions) −20 20 −20 −15 Event Day −5 0 5 Event Day 1000 500 0 0 Cum. dealer inventory (USD millions) 1500 (b) Inventory: 4 days lag 1000 2000 3000 4000 5000 (a) Volume: 4 days lag Total Daily volume (USD millions) −10 −20 −10 0 10 20 −20 −15 Event Day 0 5 10 200 400 600 800 0 Cum. dealer inventory (USD millions) −400 Total Daily volume (USD millions) 500 1000 1500 2000 2500 −10 0 (d) Inventory: 11 days lag 0 −20 −5 Event Day (c) Volume: 11 days lag −30 −10 20 Event Day −30 −20 −10 0 Event Day (e) Volume: 17 days lag (f) Inventory: 17 days lag Figure 8: Trading and inventory for specific downgrade constellations The graphs show trading activity (calculated as in Figure 2) and cumulative inventory (calculated as in Figure 4). Event time in these graphs are relative to the index exclusion date (the right vertical line). The left vertical line is the downgrade date. In each pair of graphs the time lag between downgrade date and index exclusion is kept constant at either 4, 11, or 17 days. Volume and inventory are not averaged as in the former graphs. Furthermore, these time lags have been chosen because they are the ones with the highest aggregated trading activity at the downgrade date (among all time lags). 49